A Cluster-Based Incremental Recommendation Algorithm on Stream Processing Architecture

By helping users discover books they may be interested in, recommender systems fully exploit the resources of digital libraries and better facilitate users' reading demands. Traditional memory-based collaborative filtering (CF) methods are effective and easy to interpret. However, when datasets become larger, the traditional way turns to be infeasible in both time and space. In order to address this challenge, we propose an incremental, cluster-based algorithm on Stream Processing Architecture, which is scalable and suitable to real-time environment. Our experimental results on MovieLens datasets and CADAL user-chapter logs show our algorithm is efficient, while still maintains comparable accuracy and interpretability.